Yocto Project Development Manual

Systems across a large team should meet the needs of two types of developers: those working on the contents of the operating system image itself and those developing applications. Regardless of the type of developer, their workstations must be both reasonably powerful and run Linux.

Keeping your Metadata and any software you are developing under the control of an SCM system that is compatible with the OpenEmbedded build system is advisable. Of the SCMs BitBake supports, the Yocto Project team strongly recommends using Git. Git is a distributed system that is easy to backup (each checkout is a backup in itself), allows you to work remotely, and then connects back to the infrastructure.

It is relatively easy to set up Git services and create infrastructure like http://git.yoctoproject.org, which is based on server software called gitolite with cgit being used to generate the web interface that lets you view the repositories. The gitolite software identifies users using ssh keys and allows branch-based access controls to repositories that you can control as little or as much as necessary.

Autobuilders are often the core of a development project. It is here that changes from individual developers are brought together and centrally tested and subsequent decisions about releases can be made. Autobuilders also allow for "continuous integration" style testing of software components and regression identification and tracking.

See "Welcome to the buildbot for the Yocto Project" for the Yocto Project's reference implementation that uses buildbot. The Yocto Project team has found this implementation works well in this role. A public example of this is the Yocto Project Autobuilders, which we use to test the overall health of the project.

The features of this system are:

The Yocto Project itself uses a hierarchical structure and a pull model. Scripts exist to create and send pull requests (i.e. create-pull-request and send-pull-request). This model is in line with other open source projects where maintainers are responsible for specific areas of the project and a single maintainer handles the final "top-of-tree" merges.

As with any development environment, it is important to document the policy used as well as any main project guidelines so they are understood by everyone. It is also a good idea to have well structured commit messages, which are usually a part of a project's guidelines. Good commit messages are essential when looking back in time and trying to understand why changes were made.

If you discover that changes are needed to the core layer of the project, it is worth sharing those with the community as soon as possible. Chances are if you have discovered the need for changes, someone else in the community needs them also.

This section summarizes the key recommendations described in the previous sections:

As mentioned earlier in the section "Yocto Project Source Repositories", the Yocto Project maintains source repositories at http://git.yoctoproject.org/cgit.cgi. If you look at this web-interface of the repositories, each item is a separate Git repository.

Git repositories use branching techniques that track content change (not files) within a project (e.g. a new feature or updated documentation). Creating a tree-like structure based on project divergence allows for excellent historical information over the life of a project. This methodology also allows for an environment from which you can do lots of local experimentation on projects as you develop changes or new features.

A Git repository represents all development efforts for a given project. For example, the Git repository poky contains all changes and developments for Poky over the course of its entire life. That means that all changes that make up all releases are captured. The repository maintains a complete history of changes.

You can create a local copy of any repository by "cloning" it with the Git clone command. When you clone a Git repository, you end up with an identical copy of the repository on your development system. Once you have a local copy of a repository, you can take steps to develop locally. For examples on how to clone Git repositories, see the "Getting Set Up" section.

It is important to understand that Git tracks content change and not files. Git uses "branches" to organize different development efforts. For example, the poky repository has bernard, edison, denzil, danny and master branches among others. You can see all the branches by going to http://git.yoctoproject.org/cgit.cgi/poky/ and clicking on the [...] link beneath the "Branch" heading.

Each of these branches represents a specific area of development. The master branch represents the current or most recent development. All other branches represent off-shoots of the master branch.

When you create a local copy of a Git repository, the copy has the same set of branches as the original. This means you can use Git to create a local working area (also called a branch) that tracks a specific development branch from the source Git repository. in other words, you can define your local Git environment to work on any development branch in the repository. To help illustrate, here is a set of commands that creates a local copy of the poky Git repository and then creates and checks out a local Git branch that tracks the Yocto Project 1.4.4 Release (dylan) development:

     $ cd ~
     $ git clone git://git.yoctoproject.org/poky
     $ cd poky
     $ git checkout -b dylan origin/dylan
            

In this example, the name of the top-level directory of your local Source Directory is poky, and the name of that local working area (local branch) you just created and checked out is dylan. The files in your local repository now reflect the same files that are in the dylan development branch of the Yocto Project's poky upstream repository. It is important to understand that when you create and checkout a local working branch based on a branch name, your local environment matches the "tip" of that development branch at the time you created your local branch, which could be different than the files at the time of a similarly named release. In other words, creating and checking out a local branch based on the dylan branch name is not the same as cloning and checking out the master branch. Keep reading to see how you create a local snapshot of a Yocto Project Release.

Git uses "tags" to mark specific changes in a repository. Typically, a tag is used to mark a special point such as the final change before a project is released. You can see the tags used with the poky Git repository by going to http://git.yoctoproject.org/cgit.cgi/poky/ and clicking on the [...] link beneath the "Tag" heading.

Some key tags are bernard-5.0, denzil-7.0, and dylan-9.0.3. These tags represent Yocto Project releases.

When you create a local copy of the Git repository, you also have access to all the tags. Similar to branches, you can create and checkout a local working Git branch based on a tag name. When you do this, you get a snapshot of the Git repository that reflects the state of the files when the change was made associated with that tag. The most common use is to checkout a working branch that matches a specific Yocto Project release. Here is an example:

     $ cd ~
     $ git clone git://git.yoctoproject.org/poky
     $ cd poky
     $ git checkout -b my-dylan-9.0.3 dylan-9.0.3
            

In this example, the name of the top-level directory of your local Yocto Project Files Git repository is poky. And, the name of the local branch you have created and checked out is my-dylan-9.0.3. The files in your repository now exactly match the Yocto Project 1.4.4 Release tag (dylan-9.0.3). It is important to understand that when you create and checkout a local working branch based on a tag, your environment matches a specific point in time and not the entire development branch.

Git has an extensive set of commands that lets you manage changes and perform collaboration over the life of a project. Conveniently though, you can manage with a small set of basic operations and workflows once you understand the basic philosophy behind Git. You do not have to be an expert in Git to be functional. A good place to look for instruction on a minimal set of Git commands is here. If you need to download Git, you can do so here.

If you don’t know much about Git, you should educate yourself by visiting the links previously mentioned.

The following list briefly describes some basic Git operations as a way to get started. As with any set of commands, this list (in most cases) simply shows the base command and omits the many arguments they support. See the Git documentation for complete descriptions and strategies on how to use these commands:

The basic flow for pushing a change to an upstream "contrib" Git repository is as follows:

You can find general Git information on how to push a change upstream in the Git Community Book.

You can submit patches without using the create-pull-request and send-pull-request scripts described in the previous section. However, keep in mind, the preferred method is to use the scripts.

Depending on the components changed, you need to submit the email to a specific mailing list. For some guidance on which mailing list to use, see the list in the "How to Submit a Change" section. For a description of the available mailing lists, see the "Mailing Lists" section in the Yocto Project Reference Manual.

Here is the general procedure on how to submit a patch through email without using the scripts:

A BSP is a package of recipes that, when applied during a build, results in an image that you can run on a particular board. Thus, the package when compiled into the new image, supports the operation of the board.

The remainder of this section presents the basic steps used to create a BSP using the Yocto Project's BSP Tools. Although not required for BSP creation, the meta-intel repository, which contains many BSPs supported by the Yocto Project, is part of the example.

For an example that shows how to create a new layer using the tools, see the "Creating a New BSP Layer Using the yocto-bsp Script" section in the Yocto Project Board Support Package (BSP) Developer's Guide.

The following illustration and list summarize the BSP creation general workflow.

You can view a video presentation on "Building Custom Embedded Images with Yocto" at Free Electrons. You can also find supplemental information in The Board Support Package (BSP) Development Guide. Finally, there is wiki page write up of the example also located here that you might find helpful.

Kernel modification involves changing the Yocto Project kernel, which could involve changing configuration options as well as adding new kernel recipes. Configuration changes can be added in the form of configuration fragments, while recipe modification comes through the kernel's recipes-kernel area in a kernel layer you create.

The remainder of this section presents a high-level overview of the Yocto Project kernel architecture and the steps to modify the kernel. You can reference the "Patching the Kernel" section for an example that changes the source code of the kernel. For information on how to configure the kernel, see the "Configuring the Kernel" section. For more information on the kernel and on modifying the kernel, see the Yocto Project Linux Kernel Development Manual.

4.1.2.1. Kernel Overview

Traditionally, when one thinks of a patched kernel, they think of a base kernel source tree and a fixed structure that contains kernel patches. The Yocto Project, however, employs mechanisms that, in a sense, result in a kernel source generator. By the end of this section, this analogy will become clearer.

You can find a web interface to the Yocto Project kernel source repositories at http://git.yoctoproject.org. If you look at the interface, you will see to the left a grouping of Git repositories titled "Yocto Linux Kernel." Within this group, you will find several kernels supported by the Yocto Project:

• linux-yocto-3.2 - The stable Yocto Project kernel to use with the Yocto Project Release 1.2. This kernel is based on the Linux 3.2 released kernel.

• linux-yocto-3.4 - The stable Yocto Project kernel to use with the Yocto Project Release 1.3. This kernel is based on the Linux 3.4 released kernel.

• linux-yocto-3.8 - The stable Yocto Project kernel to use with the Yocto Project Release 1.4. This kernel is based on the Linux 3.8 released kernel.

• linux-yocto-dev - A development kernel based on the latest upstream release candidate available.

The kernels are maintained using the Git revision control system that structures them using the familiar "tree", "branch", and "leaf" scheme. Branches represent diversions from general code to more specific code, while leaves represent the end-points for a complete and unique kernel whose source files, when gathered from the root of the tree to the leaf, accumulate to create the files necessary for a specific piece of hardware and its features. The following figure displays this concept:

Within the figure, the "Kernel.org Branch Point" represents the point in the tree where a supported base kernel is modified from the Linux kernel. For example, this could be the branch point for the linux-yocto-3.4 kernel. Thus, everything further to the right in the structure is based on the linux-yocto-3.4 kernel. Branch points to right in the figure represent where the linux-yocto-3.4 kernel is modified for specific hardware or types of kernels, such as real-time kernels. Each leaf thus represents the end-point for a kernel designed to run on a specific targeted device.

The overall result is a Git-maintained repository from which all the supported kernel types can be derived for all the supported devices. A big advantage to this scheme is the sharing of common features by keeping them in "larger" branches within the tree. This practice eliminates redundant storage of similar features shared among kernels.

Note

Keep in mind the figure does not take into account all the supported Yocto Project kernel types, but rather shows a single generic kernel just for conceptual purposes. Also keep in mind that this structure represents the Yocto Project source repositories that are either pulled from during the build or established on the host development system prior to the build by either cloning a particular kernel's Git repository or by downloading and unpacking a tarball.

Upstream storage of all the available kernel source code is one thing, while representing and using the code on your host development system is another. Conceptually, you can think of the kernel source repositories as all the source files necessary for all the supported kernels. As a developer, you are just interested in the source files for the kernel on which you are working. And, furthermore, you need them available on your host system.

Kernel source code is available on your host system a couple of different ways. If you are working in the kernel all the time, you probably would want to set up your own local Git repository of the kernel tree. If you just need to make some patches to the kernel, you can access temporary kernel source files that were extracted and used during a build. We will just talk about working with the temporary source code. For more information on how to get kernel source code onto your host system, see the "Yocto Project Kernel" bulleted item earlier in the manual.

What happens during the build? When you build the kernel on your development system, all files needed for the build are taken from the source repositories pointed to by the SRC_URI variable and gathered in a temporary work area where they are subsequently used to create the unique kernel. Thus, in a sense, the process constructs a local source tree specific to your kernel to generate the new kernel image - a source generator if you will.

The following figure shows the temporary file structure created on your host system when the build occurs. This Build Directory contains all the source files used during the build.

Again, for additional information the Yocto Project kernel's architecture and its branching strategy, see the Yocto Project Linux Kernel Development Manual. You can also reference the "Patching the Kernel" section for a detailed example that modifies the kernel.

4.1.2.2. Kernel Modification Workflow

This illustration and the following list summarizes the kernel modification general workflow.

1. Set up your host development system to support development using the Yocto Project: See "The Linux Distribution" and "The Packages" sections both in the Yocto Project Quick Start for requirements.

2. Establish a local copy of project files on your system: Having the Source Directory on your system gives you access to the build process and tools you need. For information on how to get these files, see the bulleted item "Yocto Project Release" earlier in this manual.

3. Establish the temporary kernel source files: Temporary kernel source files are kept in the Build Directory created by the OpenEmbedded build system when you run BitBake. If you have never built the kernel you are interested in, you need to run an initial build to establish local kernel source files.

   If you are building an image for the first time, you need to get the build environment ready by sourcing the environment setup script. You also need to be sure two key configuration files (local.conf and bblayers.conf) are configured appropriately.

   The entire process for building an image is overviewed in the "Building an Image" section of the Yocto Project Quick Start. You might want to reference this information. You can find more information on BitBake in the user manual, which is found in the bitbake/doc/manual directory of the Source Directory.

   The build process supports several types of images to satisfy different needs. See the "Images" chapter in the Yocto Project Reference Manual for information on supported images.

4. Make changes to the kernel source code if applicable: Modifying the kernel does not always mean directly changing source files. However, if you have to do this, you make the changes to the files in the Build directory.

5. Make kernel configuration changes if applicable: If your situation calls for changing the kernel's configuration, you can use the yocto-kernel script or menuconfig to enable and disable kernel configurations. Using the script lets you interactively set up kernel configurations. Using menuconfig allows you to interactively develop and test the configuration changes you are making to the kernel. When saved, changes using menuconfig update the kernel's .config file. Try to resist the temptation of directly editing the .config file found in the Build Directory at tmp/sysroots/<machine-name>/kernel. Doing so, can produce unexpected results when the OpenEmbedded build system regenerates the configuration file.

   Once you are satisfied with the configuration changes made using menuconfig, you can directly compare the .config file against a saved original and gather those changes into a config fragment to be referenced from within the kernel's .bbappend file.

6. Rebuild the kernel image with your changes: Rebuilding the kernel image applies your changes.

To help you understand how application development works using the ADT, this section provides an overview of the general development process and a detailed example of the process as it is used from within the Eclipse IDE.

The following illustration and list summarize the application development general workflow.

The Eclipse IDE is a popular development environment and it fully supports development using the Yocto Project.

When you install and configure the Eclipse Yocto Project Plug-in into the Eclipse IDE, you maximize your Yocto Project experience. Installing and configuring the Plug-in results in an environment that has extensions specifically designed to let you more easily develop software. These extensions allow for cross-compilation, deployment, and execution of your output into a QEMU emulation session as well as actual target hardware. You can also perform cross-debugging and profiling. The environment also supports a suite of tools that allows you to perform remote profiling, tracing, collection of power data, collection of latency data, and collection of performance data.

This section describes how to install and configure the Eclipse IDE Yocto Plug-in and how to use it to develop your application.

     $ cd ~
     $ tar -xzvf ~/Downloads/eclipse-SDK-4.2-linux-gtk-x86_64.tar.gz
                    

     Java.lang.OutOfMemoryError: PermGen space
                    

     eclipse --vmargs --XX:PermSize=256M
                    

If you want to develop an application without prior installation of the ADT, you still can employ the Cross Development Toolchain, the QEMU emulator, and a number of supported target image files. You just need to follow these general steps:

During a build, the unpacked temporary source code used by recipes to build packages is available in the Build Directory as defined by the S variable. Below is the default value for the S variable as defined in the meta/conf/bitbake.conf configuration file in the Source Directory:

     S = ${WORKDIR}/${BP}
            

You should be aware that many recipes override the S variable. For example, recipes that fetch their source from Git usually set S to ${WORKDIR}/git.

     BP = ${BPN}-${PV}
                

The path to the work directory for the recipe (WORKDIR) depends on the recipe name and the architecture of the target device. For example, here is the work directory for recipes and resulting packages that are not device-dependent:

     ${TMPDIR}/work/${PACKAGE_ARCH}-poky-${TARGET_OS}/${PN}-${PV}-${PR}
            

Let's look at an example without variables. Assuming a top-level Source Directory named poky and a default Build Directory of poky/build, the following is the work directory for the acl recipe that creates the acl package:

     ~/poky/build/tmp/work/i586-poky-linux/acl-2.2.51-r3
            

If your resulting package is dependent on the target device, the work directory varies slightly:

     ${TMPDIR}/work/${MACHINE}-poky-${TARGET_OS}/${PN}-${PV}-${PR}
            

Again, assuming top-level Source Directory named poky and a default Build Directory of poky/build, the following are the work and temporary source directories, respectively, for the acl package that is being built for a MIPS-based device:

     ~/poky/build/tmp/work/mips-poky-linux/acl-2.2.51-r2
     ~/poky/build/tmp/work/mips-poky-linux/acl-2.2.51-r2/acl-2.2.51
            

Now that you know where to locate the directory that has the temporary source code, you can use a Quilt or Git workflow to make your edits, test the changes, and preserve the changes in the form of patches.

Quilt is a powerful tool that allows you to capture source code changes without having a clean source tree. This section outlines the typical workflow you can use to modify temporary source code, test changes, and then preserve the changes in the form of a patch all using Quilt.

Follow these general steps:

Git is an even more powerful tool that allows you to capture source code changes without having a clean source tree. This section outlines the typical workflow you can use to modify temporary source code, test changes, and then preserve the changes in the form of a patch all using Git. For general information on Git as it is used in the Yocto Project, see the "Git" section.

Follow these general steps:

The Source Directory contains both general layers and BSP layers right out of the box. You can easily identify layers that ship with a Yocto Project release in the Source Directory by their folder names. Folders that are layers begin with the string meta.

For example, when you set up the Source Directory structure, you will see several layers: meta, meta-hob, meta-skeleton, meta-yocto, and meta-yocto-bsp. Each of these folders is a layer.

Furthermore, if you set up a local copy of the meta-intel Git repository and then explore the folder of that general layer, you will discover many BSP layers inside. For more information on BSP layers, see the "BSP Layers" section in the Yocto Project Board Support Package (BSP) Developer's Guide.

It is very easy to create your own layers to use with the OpenEmbedded build system. The Yocto Project ships with scripts that speed up creating general layers and BSP layers. This section describes the steps you perform by hand to create a layer so that you can better understand them. For information about the layer-creation scripts, see the "Creating a New BSP Layer Using the yocto-bsp Script" section in the Yocto Project Board Support Package (BSP) Developer's Guide and the "Creating a General Layer Using the yocto-layer Script" section further down in this manual.

Follow these general steps to create your layer:

To create layers that are easier to maintain, you should consider the following:

We also recommend the following:

Following these recommendations keeps your Source Directory and its configuration entirely inside the Yocto Project's core base.

Before the OpenEmbedded build system can use your new layer, you need to enable it. To enable your layer, simply add your layer's path to the BBLAYERS variable in your conf/bblayers.conf file, which is found in the Build Directory. The following example shows how to enable a layer named meta-mylayer:

     LCONF_VERSION = "6"

     BBPATH = "${TOPDIR}"
     BBFILES ?= ""

     BBLAYERS ?= " \
       $HOME/poky/meta \
       $HOME/poky/meta-yocto \
       $HOME/poky/meta-yocto-bsp \
       $HOME/poky/meta-mylayer \
       "

     BBLAYERS_NON_REMOVABLE ?= " \
       $HOME/poky/meta \
       $HOME/poky/meta-yocto \
       "
                

BitBake parses each conf/layer.conf file as specified in the BBLAYERS variable within the conf/bblayers.conf file. During the processing of each conf/layer.conf file, BitBake adds the recipes, classes and configurations contained within the particular layer to the source directory.

Recipes used to append Metadata to other recipes are called BitBake append files. BitBake append files use the .bbappend file type suffix, while the corresponding recipes to which Metadata is being appended use the .bb file type suffix.

A .bbappend file allows your layer to make additions or changes to the content of another layer's recipe without having to copy the other recipe into your layer. Your .bbappend file resides in your layer, while the main .bb recipe file to which you are appending Metadata resides in a different layer.

Append files must have the same root names as their corresponding recipes. For example, the append file someapp_1.4.4.bbappend must apply to someapp_1.4.4.bb. This means the original recipe and append file names are version number-specific. If the corresponding recipe is renamed to update to a newer version, the corresponding .bbappend file must be renamed as well. During the build process, BitBake displays an error on starting if it detects a .bbappend file that does not have a corresponding recipe with a matching name. See the BB_DANGLINGAPPENDS_WARNONLY variable for information on how to handle this error.

Being able to append information to an existing recipe not only avoids duplication, but also automatically applies recipe changes in a different layer to your layer. If you were copying recipes, you would have to manually merge changes as they occur.

As an example, consider the main formfactor recipe and a corresponding formfactor append file both from the Source Directory. Here is the main formfactor recipe, which is named formfactor_0.0.bb and located in the "meta" layer at meta/recipes-bsp/formfactor:

     DESCRIPTION = "Device formfactor information"
     SECTION = "base"
     LICENSE = "MIT"
     LIC_FILES_CHKSUM = "file://${COREBASE}/LICENSE;md5=3f40d7994397109285ec7b81fdeb3b58 \
                         file://${COREBASE}/meta/COPYING.MIT;md5=3da9cfbcb788c80a0384361b4de20420"
     PR = "r21"

     SRC_URI = "file://config file://machconfig"
     S = "${WORKDIR}"

     PACKAGE_ARCH = "${MACHINE_ARCH}"
     INHIBIT_DEFAULT_DEPS = "1"

     do_install() {
     	# Only install file if it has a contents
             install -d ${D}${sysconfdir}/formfactor/
             install -m 0644 ${S}/config ${D}${sysconfdir}/formfactor/
     	if [ -s "${S}/machconfig" ]; then
     	        install -m 0644 ${S}/machconfig ${D}${sysconfdir}/formfactor/
     	fi
     }          

In the main recipe, note the SRC_URI variable, which tells the OpenEmbedded build system where to find files during the build.

Following is the append file, which is named formfactor_0.0.bbappend and is from the Crown Bay BSP Layer named meta-intel/meta-crownbay. The file is in recipes-bsp/formfactor:

     FILESEXTRAPATHS_prepend := "${THISDIR}/${PN}:"

     PRINC := "${@int(PRINC) + 2}"
                

By default, the build system uses the FILESPATH variable to locate files. This append file extends the locations by setting the FILESEXTRAPATHS variable. Setting this variable in the .bbappend file is the most reliable and recommended method for adding directories to the search path used by the build system to find files.

The statement in this example extends the directories to include ${THISDIR}/${PN}, which resolves to a directory named formfactor in the same directory in which the append file resides (i.e. meta-intel/meta-crownbay/recipes-bsp/formfactor/formfactor. This implies that you must have the supporting directory structure set up that will contain any files or patches you will be including from the layer.

Using the immediate expansion assignment operator := is important because of the reference to THISDIR. The trailing colon character is important as it ensures that items in the list remain colon-separated.

Each layer is assigned a priority value. Priority values control which layer takes precedence if there are recipe files with the same name in multiple layers. For these cases, the recipe file from the layer with a higher priority number takes precedence. Priority values also affect the order in which multiple .bbappend files for the same recipe are applied. You can either specify the priority manually, or allow the build system to calculate it based on the layer's dependencies.

To specify the layer's priority manually, use the BBFILE_PRIORITY variable. For example:

     BBFILE_PRIORITY_mylayer = "1"
                

You can use the BitBake layer management tool to provide a view into the structure of recipes across a multi-layer project. Being able to generate output that reports on configured layers with their paths and priorities and on .bbappend files and their applicable recipes can help to reveal potential problems.

Use the following form when running the layer management tool.

     $ bitbake-layers <command> [arguments]
                

The following list describes the available commands:

The yocto-layer script simplifies creating a new general layer.

The default mode of the script's operation is to prompt you for information needed to generate the layer:

Use the yocto-layer create sub-command to create a new general layer. In its simplest form, you can create a layer as follows:

     $ yocto-layer create mylayer
                

The previous example creates a layer named meta-mylayer in the current directory.

As the yocto-layer create command runs, default values for the prompts appear in brackets. Pressing enter without supplying anything for the prompts or pressing enter and providing an invalid response causes the script to accept the default value. Once the script completes, the new layer is created in the current working directory. The script names the layer by prepending meta- to the name you provide.

Minimally, the script creates the following within the layer:

If you choose to generate a sample recipe file, the script prompts you for the name for the recipe and then creates it in <layer>/recipes-example/example/. in a directory named recipes-example. The script creates a .bb file and a directory, which contains a sample helloworld.c source file and along with a sample patch file. If you do not provide a recipe name, the script uses "example".

If you choose to generate a sample append file, the script prompts you for the name for the file and then creates it in <layer>/recipes-example-bbappend/example-bbappend/. The script creates a .bbappend file and a directory, which contains a sample patch file. If you do not provide a recipe name, the script uses "example". The script also prompts you for the version of the append file. The version should match the recipe to which the append file is associated.

The easiest way to see how the yocto-layer script works is to experiment with the script. You can also read the usage information by entering the following:

     $ yocto-layer help
                

Once you create your general layer, you must add it to your bblayers.conf file. Here is an example:

     BBLAYERS = ?" \
        /usr/local/src/yocto/meta \
        /usr/local/src/yocto/meta-yocto \
        /usr/local/src/yocto/meta-yocto-bsp \
        /usr/local/src/yocto/meta-mylayer \
        "

     BBLAYERS_NON_REMOVABLE ?= " \
        /usr/local/src/yocto/meta \
        /usr/local/src/yocto/meta-yocto \
        "
                

Adding the layer to this file enables the build system to locate the layer during the build.

One way to get additional software into an image is to create a custom image. The following example shows the form for the two lines you need:

     IMAGE_INSTALL = "packagegroup-core-x11-base package1 package2"

     inherit core-image
                

By creating a custom image, a developer has total control over the contents of the image. It is important to use the correct names of packages in the IMAGE_INSTALL variable. You must use the OpenEmbedded notation and not the Debian notation for the names (e.g. eglibc-dev instead of libc6-dev).

The other method for creating a custom image is to base it on an existing image. For example, if you want to create an image based on core-image-sato but add the additional package strace to the image, copy the meta/recipes-sato/images/core-image-sato.bb to a new .bb and add the following line to the end of the copy:

     IMAGE_INSTALL += "strace"
                

For complex custom images, the best approach is to create a custom package group recipe that is used to build the image or images. A good example of a package group recipe is meta/recipes-core/packagegroups/packagegroup-core-boot.bb. The PACKAGES variable lists the package group packages you wish to produce. inherit packagegroup sets appropriate default values and automatically adds -dev and -dbg complementary packages for every package specified in PACKAGES. Note that the inherit line should be towards the top of the recipe, certainly before you set PACKAGES. For each package you specify in PACKAGES, you can use RDEPENDS and RRECOMMENDS entries to provide a list of packages the parent task package should contain. Following is an example:

     DESCRIPTION = "My Custom Package Groups"

     inherit packagegroup

     PACKAGES = "\
         packagegroup-custom-apps \
         packagegroup-custom-tools \
         "

     RDEPENDS_packagegroup-custom-apps = "\
         dropbear \
         portmap \
         psplash"

     RDEPENDS_packagegroup-custom-tools = "\
         oprofile \
         oprofileui-server \
         lttng-control \
         lttng-viewer"

     RRECOMMENDS_packagegroup-custom-tools = "\
         kernel-module-oprofile"
                

In the previous example, two package group packages are created with their dependencies and their recommended package dependencies listed: packagegroup-custom-apps, and packagegroup-custom-tools. To build an image using these package group packages, you need to add packagegroup-custom-apps and/or packagegroup-custom-tools to IMAGE_INSTALL. For other forms of image dependencies see the other areas of this section.

You might want to customize your image by enabling or disabling high-level image features by using the IMAGE_FEATURES and EXTRA_IMAGE_FEATURES variables. Although the functions for both variables are nearly equivalent, best practices dictate using IMAGE_FEATURES from within a recipe and using EXTRA_IMAGE_FEATURES from within your local.conf file, which is found in the Build Directory.

To understand how these features work, the best reference is meta/classes/core-image.bbclass. In summary, the file looks at the contents of the IMAGE_FEATURES variable and then maps those contents into a set of package groups. Based on this information, the build system automatically adds the appropriate packages to the IMAGE_INSTALL variable. Effectively, you are enabling extra features by extending the class or creating a custom class for use with specialized image .bb files.

Use the EXTRA_IMAGE_FEATURES variable from within your local configuration file. Using a separate area from which to enable features with this variable helps you avoid overwriting the features in the image recipe that are enabled with IMAGE_FEATURES. The value of EXTRA_IMAGE_FEATURES is added to IMAGE_FEATURES within meta/conf/bitbake.conf.

To illustrate how you can use these variables to modify your image, consider an example that selects the SSH server. The Yocto Project ships with two SSH servers you can use with your images: Dropbear and OpenSSH. Dropbear is a minimal SSH server appropriate for resource-constrained environments, while OpenSSH is a well-known standard SSH server implementation. By default, the core-image-sato image is configured to use Dropbear. The core-image-basic and core-image-lsb images both include OpenSSH. The core-image-minimal image does not contain an SSH server.

You can customize your image and change these defaults. Edit the IMAGE_FEATURES variable in your recipe or use the EXTRA_IMAGE_FEATURES in your local.conf file so that it configures the image you are working with to include ssh-server-dropbear or ssh-server-openssh.

It is possible to customize image contents by using variables from your local configuration in your conf/local.conf file. Because it is limited to local use, this method generally only allows you to add packages and is not as flexible as creating your own customized image. When you add packages using local variables this way, you need to realize that these variable changes affect all images at the same time and might not be what you require.

The simplest way to add extra packages to all images is by using the IMAGE_INSTALL variable with the _append operator:

     IMAGE_INSTALL_append = " strace"
                

Use of the syntax is important. Specifically, the space between the quote and the package name, which is strace in this example. This space is required since the _append operator does not add the space.

Furthermore, you must use _append instead of the += operator if you want to avoid ordering issues. The reason for this is because doing so unconditionally appends to the variable and avoids ordering problems due to the variable being set in image recipes and .bbclass files with operators like ?=. Using _append ensures the operation takes affect.

As shown in its simplest use, IMAGE_INSTALL_append affects all images. It is possible to extend the syntax so that the variable applies to a specific image only. Here is an example:

     IMAGE_INSTALL_append_pn-core-image-minimal = " strace"
                

This example adds strace to core-image-minimal only.

You can add packages using a similar approach through the CORE_IMAGE_EXTRA_INSTALL variable. If you use this variable, only core-image-* images are affected.

Building an application from a single file that is stored locally (e.g. under files/) requires a recipe that has the file listed in the SRC_URI variable. Additionally, you need to manually write the do_compile and do_install tasks. The S variable defines the directory containing the source code, which is set to WORKDIR in this case - the directory BitBake uses for the build.

     DESCRIPTION = "Simple helloworld application"
     SECTION = "examples"
     LICENSE = "MIT"
     LIC_FILES_CHKSUM = "file://${COMMON_LICENSE_DIR}/MIT;md5=0835ade698e0bcf8506ecda2f7b4f302"
     PR = "r0"

     SRC_URI = "file://helloworld.c"

     S = "${WORKDIR}"

     do_compile() {
     	${CC} helloworld.c -o helloworld
     }

     do_install() {
     	install -d ${D}${bindir}
     	install -m 0755 helloworld ${D}${bindir}
     }
                

By default, the helloworld, helloworld-dbg, and helloworld-dev packages are built. For information on how to customize the packaging process, see the "Splitting an Application into Multiple Packages" section.

Applications that use Autotools such as autoconf and automake require a recipe that has a source archive listed in SRC_URI and also inherits Autotools, which instructs BitBake to use the autotools.bbclass file, which contains the definitions of all the steps needed to build an Autotool-based application. The result of the build is automatically packaged. And, if the application uses NLS for localization, packages with local information are generated (one package per language). Following is one example: (hello_2.3.bb)

     DESCRIPTION = "GNU Helloworld application"
     SECTION = "examples"
     LICENSE = "GPLv2+"
     LIC_FILES_CHKSUM = "file://COPYING;md5=751419260aa954499f7abaabaa882bbe"
     PR = "r0"

     SRC_URI = "${GNU_MIRROR}/hello/hello-${PV}.tar.gz"

     inherit autotools gettext
                 

The variable LIC_FILES_CHKSUM is used to track source license changes as described in the "Tracking License Changes" section. You can quickly create Autotool-based recipes in a manner similar to the previous example.

Applications that use GNU make also require a recipe that has the source archive listed in SRC_URI. You do not need to add a do_compile step since by default BitBake starts the make command to compile the application. If you need additional make options, you should store them in the EXTRA_OEMAKE variable. BitBake passes these options into the make GNU invocation. Note that a do_install task is still required. Otherwise, BitBake runs an empty do_install task by default.

Some applications might require extra parameters to be passed to the compiler. For example, the application might need an additional header path. You can accomplish this by adding to the CFLAGS variable. The following example shows this:

     CFLAGS_prepend = "-I ${S}/include "
                

In the following example, mtd-utils is a makefile-based package:

     DESCRIPTION = "Tools for managing memory technology devices."
     SECTION = "base"
     DEPENDS = "zlib lzo e2fsprogs util-linux"
     HOMEPAGE = "http://www.linux-mtd.infradead.org/"
     LICENSE = "GPLv2+"
     LIC_FILES_CHKSUM = "file://COPYING;md5=0636e73ff0215e8d672dc4c32c317bb3 \
                    file://include/common.h;beginline=1;endline=17;md5=ba05b07912a44ea2bf81ce409380049c"

     SRC_URI = "git://git.infradead.org/mtd-utils.git;protocol=git;tag=995cfe51b0a3cf32f381c140bf72b21bf91cef1b \
	     	file://add-exclusion-to-mkfs-jffs2-git-2.patch"

     S = "${WORKDIR}/git/"

     PR = "r1"

     EXTRA_OEMAKE = "'CC=${CC}' 'RANLIB=${RANLIB}' 'AR=${AR}' \
        'CFLAGS=${CFLAGS} -I${S}/include -DWITHOUT_XATTR' 'BUILDDIR=${S}'"

     do_install () {
	     oe_runmake install DESTDIR=${D} SBINDIR=${sbindir} MANDIR=${mandir} \
            INCLUDEDIR=${includedir}
	     install -d ${D}${includedir}/mtd/
	     for f in ${S}/include/mtd/*.h; do
	     	install -m 0644 $f ${D}${includedir}/mtd/
	     done
     }

     PARALLEL_MAKE = ""

     BBCLASSEXTEND = "native"
                

If your sources are available as a tarball instead of a Git repository, you will need to provide the URL to the tarball as well as an md5 or sha256 sum of the download. Here is an example:

     SRC_URI="ftp://ftp.infradead.org/pub/mtd-utils/mtd-utils-1.4.9.tar.bz2"
     SRC_URI[md5sum]="82b8e714b90674896570968f70ca778b"
                

You can generate the md5 or sha256 sums by using the md5sum or sha256sum commands with the target file as the only argument. Here is an example:

     $ md5sum mtd-utils-1.4.9.tar.bz2
     82b8e714b90674896570968f70ca778b mtd-utils-1.4.9.tar.bz2
                

You can use the variables PACKAGES and FILES to split an application into multiple packages.

Following is an example that uses the libXpm recipe. By default, this recipe generates a single package that contains the library along with a few binaries. You can modify the recipe to split the binaries into separate packages:

     require xorg-lib-common.inc

     DESCRIPTION = "X11 Pixmap library"
     LICENSE = "X-BSD"
     LIC_FILES_CHKSUM = "file://COPYING;md5=3e07763d16963c3af12db271a31abaa5"
     DEPENDS += "libxext libsm libxt"
     PR = "r3"
     PE = "1"

     XORG_PN = "libXpm"

     PACKAGES =+ "sxpm cxpm"
     FILES_cxpm = "${bindir}/cxpm"
     FILES_sxpm = "${bindir}/sxpm"
                

In the previous example, we want to ship the sxpm and cxpm binaries in separate packages. Since bindir would be packaged into the main PN package by default, we prepend the PACKAGES variable so additional package names are added to the start of list. This results in the extra FILES_* variables then containing information that define which files and directories go into which packages. Files included by earlier packages are skipped by latter packages. Thus, the main PN package does not include the above listed files.

To add a post-installation script to a package, add a pkg_postinst_PACKAGENAME() function to the .bb file and use PACKAGENAME as the name of the package you want to attach to the postinst script. Normally, PN can be used, which automatically expands to PACKAGENAME. A post-installation function has the following structure:

     pkg_postinst_PACKAGENAME () {
     #!/bin/sh -e
     # Commands to carry out
     }
                

The script defined in the post-installation function is called when the root filesystem is created. If the script succeeds, the package is marked as installed. If the script fails, the package is marked as unpacked and the script is executed when the image boots again.

Sometimes it is necessary for the execution of a post-installation script to be delayed until the first boot. For example, the script might need to be executed on the device itself. To delay script execution until boot time, use the following structure in the post-installation script:

     pkg_postinst_PACKAGENAME () {
     #!/bin/sh -e
     if [ x"$D" = "x" ]; then
          # Actions to carry out on the device go here
     else
          exit 1
     fi
     }
                

The previous example delays execution until the image boots again because the D variable points to the directory containing the image when the root filesystem is created at build time but is unset when executed on the first boot.

To add a machine configuration, you need to add a .conf file with details of the device being added to the conf/machine/ file. The name of the file determines the name the OpenEmbedded build system uses to reference the new machine.

The most important variables to set in this file are as follows:

You might also need these variables:

You can find full details on these variables in the reference section. You can leverage many existing machine .conf files from meta/conf/machine/.

The OpenEmbedded build system needs to be able to build a kernel for the machine. You need to either create a new kernel recipe for this machine, or extend an existing recipe. You can find several kernel examples in the Source Directory at meta/recipes-kernel/linux that you can use as references.

If you are creating a new recipe, normal recipe-writing rules apply for setting up a SRC_URI. Thus, you need to specify any necessary patches and set S to point at the source code. You need to create a configure task that configures the unpacked kernel with a defconfig. You can do this by using a make defconfig command or, more commonly, by copying in a suitable defconfig file and and then running make oldconfig. By making use of inherit kernel and potentially some of the linux-*.inc files, most other functionality is centralized and the the defaults of the class normally work well.

If you are extending an existing kernel, it is usually a matter of adding a suitable defconfig file. The file needs to be added into a location similar to defconfig files used for other machines in a given kernel. A possible way to do this is by listing the file in the SRC_URI and adding the machine to the expression in COMPATIBLE_MACHINE:

     COMPATIBLE_MACHINE = '(qemux86|qemumips)'
                

A formfactor configuration file provides information about the target hardware for which the image is being built and information that the build system cannot obtain from other sources such as the kernel. Some examples of information contained in a formfactor configuration file include framebuffer orientation, whether or not the system has a keyboard, the positioning of the keyboard in relation to the screen, and the screen resolution.

The build system uses reasonable defaults in most cases. However, if customization is necessary, you need to create a machconfig file in the meta/recipes-bsp/formfactor/files directory. This directory contains directories for specific machines such as qemuarm and qemux86. For information about the settings available and the defaults, see the meta/recipes-bsp/formfactor/files/config file found in the same area.

Following is an example for qemuarm:

     HAVE_TOUCHSCREEN=1
     HAVE_KEYBOARD=1

     DISPLAY_CAN_ROTATE=0
     DISPLAY_ORIENTATION=0
     #DISPLAY_WIDTH_PIXELS=640
     #DISPLAY_HEIGHT_PIXELS=480
     #DISPLAY_BPP=16
     DISPLAY_DPI=150
     DISPLAY_SUBPIXEL_ORDER=vrgb
                

If you are building a library and the library offers static linking, you can control which static library files (*.a files) get included in the built library.

The PACKAGES and FILES_* variables in the meta/conf/bitbake.conf configuration file define how files installed by the do_install task are packaged. By default, the PACKAGES variable contains ${PN}-staticdev, which includes all static library files.

Following, is part of the BitBake configuration file. You can see where the static library files are defined:

     PACKAGES = "${PN}-dbg ${PN} ${PN}-doc ${PN}-dev ${PN}-staticdev ${PN}-locale"
     PACKAGES_DYNAMIC = "${PN}-locale-*"
     FILES = ""

     FILES_${PN} = "${bindir}/* ${sbindir}/* ${libexecdir}/* ${libdir}/lib*${SOLIBS} \
                 ${sysconfdir} ${sharedstatedir} ${localstatedir} \
                 ${base_bindir}/* ${base_sbindir}/* \
                 ${base_libdir}/*${SOLIBS} \
                 ${datadir}/${BPN} ${libdir}/${BPN}/* \
                 ${datadir}/pixmaps ${datadir}/applications \
                 ${datadir}/idl ${datadir}/omf ${datadir}/sounds \
                 ${libdir}/bonobo/servers"

     FILES_${PN}-doc = "${docdir} ${mandir} ${infodir} ${datadir}/gtk-doc \
                 ${datadir}/gnome/help"
     SECTION_${PN}-doc = "doc"

     FILES_${PN}-dev = "${includedir} ${libdir}/lib*${SOLIBSDEV} ${libdir}/*.la \
                     ${libdir}/*.o ${libdir}/pkgconfig ${datadir}/pkgconfig \
                     ${datadir}/aclocal ${base_libdir}/*.o"
     SECTION_${PN}-dev = "devel"
     ALLOW_EMPTY_${PN}-dev = "1"
     RDEPENDS_${PN}-dev = "${PN} (= ${EXTENDPKGV})"

     FILES_${PN}-staticdev = "${libdir}/*.a ${base_libdir}/*.a"
     SECTION_${PN}-staticdev = "devel"
     RDEPENDS_${PN}-staticdev = "${PN}-dev (= ${EXTENDPKGV})"
                

The build system offers the ability to build libraries with different target optimizations or architecture formats and combine these together into one system image. You can link different binaries in the image against the different libraries as needed for specific use cases. This feature is called "Multilib."

An example would be where you have most of a system compiled in 32-bit mode using 32-bit libraries, but you have something large, like a database engine, that needs to be a 64-bit application and uses 64-bit libraries. Multilib allows you to get the best of both 32-bit and 64-bit libraries.

While the Multilib feature is most commonly used for 32 and 64-bit differences, the approach the build system uses facilitates different target optimizations. You could compile some binaries to use one set of libraries and other binaries to use other different sets of libraries. The libraries could differ in architecture, compiler options, or other optimizations.

This section overviews the Multilib process only. For more details on how to implement Multilib, see the Multilib wiki page.

Aside from this wiki page, several examples exist in the meta-skeleton layer found in the Source Directory:

     MACHINE = "qemux86-64"
     require conf/multilib.conf
     MULTILIBS = "multilib:lib32"
     DEFAULTTUNE_virtclass-multilib-lib32 = "x86"
     IMAGE_INSTALL = "lib32-connman"
                    

     $ bitbake core-image-sato
                    

     $  bitbake lib32-connman
                    

Situations can exist where you need to install and use multiple versions of the same library on the same system at the same time. These situations almost always exist when a library API changes and you have multiple pieces of software that depend on the separate versions of the library. To accommodate these situations, you can install multiple versions of the same library in parallel on the same system.

The process is straight forward as long as the libraries use proper versioning. With properly versioned libraries, all you need to do to individually specify the libraries is create separate, appropriately named recipes where the PN part of the name includes a portion that differentiates each library version (e.g.the major part of the version number). Thus, instead of having a single recipe that loads one version of a library (e.g. clutter), you provide multiple recipes that result in different versions of the libraries you want. As an example, the following two recipes would allow the two separate versions of the clutter library to co-exist on the same system:

     clutter-1.6_1.6.20.bb
     clutter-1.8_1.8.4.bb
                

Additionally, if you have other recipes that depend on a given library, you need to use the DEPENDS variable to create the dependency. Continuing with the same example, if you want to have a recipe depend on the 1.8 version of the clutter library, use the following in your recipe:

     DEPENDS = "clutter-1.8"
                

The easiest way to define kernel configurations is to set them through the menuconfig tool. This tool provides an interactive method with which to set kernel configurations. For general information on menuconfig, see http://en.wikipedia.org/wiki/Menuconfig.

To use the menuconfig tool in the Yocto Project development environment, you must build the tool using BitBake. Thus, the environment must be set up using the oe-init-build-env script found in the Build Directory. The following commands build and invoke menuconfig assuming the Source Directory top-level folder is ~/poky:

     $ cd ~/poky
     $ source oe-init-build-env
     $ bitbake linux-yocto -c menuconfig
                

Once menuconfig comes up, its standard interface allows you to interactively examine and configure all the kernel configuration parameters. After making your changes, simply exit the tool and save your changes to create an updated version of the .config configuration file.

Consider an example that configures the linux-yocto-3.4 kernel. The OpenEmbedded build system recognizes this kernel as linux-yocto. Thus, the following commands from the shell in which you previously sourced the environment initialization script cleans the shared state cache and the WORKDIR directory and then builds and launches menuconfig:

     $ bitbake linux-yocto -c menuconfig
                

Once menuconfig launches, use the interface to navigate through the selections to find the configuration settings in which you are interested. For example, consider the CONFIG_SMP configuration setting. You can find it at Processor Type and Features under the configuration selection Symmetric Multi-processing Support. After highlighting the selection, use the arrow keys to select or deselect the setting. When you are finished with all your selections, exit out and save them.

Saving the selections updates the .config configuration file. This is the file that the OpenEmbedded build system uses to configure the kernel during the build. You can find and examine this file in the Build Directory in tmp/work/. The actual .config is located in the area where the specific kernel is built. For example, if you were building a Linux Yocto kernel based on the Linux 3.4 kernel and you were building a QEMU image targeted for x86 architecture, the .config file would be located here:

     ~/poky/build/tmp/work/qemux86-poky-linux/linux-yocto-3.4.11+git1+84f...
        ...656ed30-r1/linux-qemux86-standard-build
                

Within the .config file, you can see the kernel settings. For example, the following entry shows that symmetric multi-processor support is not set:

     # CONFIG_SMP is not set
                

A good method to isolate changed configurations is to use a combination of the menuconfig tool and simple shell commands. Before changing configurations with menuconfig, copy the existing .config and rename it to something else, use menuconfig to make as many changes an you want and save them, then compare the renamed configuration file against the newly created file. You can use the resulting differences as your base to create configuration fragments to permanently save in your kernel layer.

Configuration fragments are simply kernel options that appear in a file placed where the OpenEmbedded build system can find and apply them. Syntactically, the configuration statement is identical to what would appear in the .config file, which is in the Build Directory in tmp/work/<arch>-poky-linux/linux-yocto-<release-specific-string>/linux-<arch>-<build-type>.

It is simple to create a configuration fragment. For example, issuing the following from the shell creates a configuration fragment file named my_smp.cfg that enables multi-processor support within the kernel:

     $ echo "CONFIG_SMP=y" >> my_smp.cfg
                

Where do you put your configuration files? You can place these configuration files in the same area pointed to by SRC_URI. The OpenEmbedded build system will pick up the configuration and add it to the kernel's configuration. For example, suppose you had a set of configuration options in a file called myconfig.cfg. If you put that file inside a directory named /linux-yocto that resides in the same directory as the kernel's append file and then add a SRC_URI statement such as the following to the kernel's append file, those configuration options will be picked up and applied when the kernel is built.

     SRC_URI += "file://myconfig.cfg"
                

As mentioned earlier, you can group related configurations into multiple files and name them all in the SRC_URI statement as well. For example, you could group separate configurations specifically for Ethernet and graphics into their own files and add those by using a SRC_URI statement like the following in your append file:

     SRC_URI += "file://myconfig.cfg \
            file://eth.cfg \
            file://gfx.cfg"
                

You can make sure the .config file is as lean or efficient as possible by reading the output of the kernel configuration fragment audit, noting any issues, making changes to correct the issues, and then repeating.

As part of the kernel build process, the kernel_configcheck task runs. This task validates the kernel configuration by checking the final .config file against the input files. During the check, the task produces warning messages for the following issues:

For each output warning, a message points to the file that contains a list of the options and a pointer to the config fragment that defines them. Collectively, the files are the key to streamlining the configuration.

To streamline the configuration, do the following:

Iteratively working through steps two through four eventually yields a minimal, streamlined configuration file. Once you have the best .config, you can build the Linux Yocto kernel.

The first step is to create a layer so you can isolate your changes:

     $cd ~/poky
     $mkdir meta-mylayer
                

Creating a directory that follows the Yocto Project layer naming conventions sets up the layer for your changes. The layer is where you place your configuration files, append files, and patch files. To learn more about creating a layer and filling it with the files you need, see the "Understanding and Creating Layers" section.

Each time you build a kernel image, the kernel source code is fetched and unpacked into the following directory:

     ${S}/linux
                

See the "Finding the Temporary Source Code" section and the S variable for more information about where source is kept during a build.

For this example, we are going to patch the init/calibrate.c file by adding some simple console printk statements that we can see when we boot the image using QEMU.

Two methods exist by which you can create the patch: Git workflow and Quilt workflow. For kernel patches, the Git workflow is more appropriate. This section assumes the Git workflow and shows the steps specific to this example.

These steps get your layer set up for the build:

Do the following to make sure the build parameters are set up for the example. Once you set up these build parameters, they do not have to change unless you change the target architecture of the machine you are building:

The following steps build your modified kernel image:

These steps boot the image and allow you to see the changes

The following list presents the overall steps you need to consider and perform to create distributions with smaller root filesystems, faster boot times, maintain your critical functionality, and avoid initial RAM disks:

Before you can reach your destination, you need to know where you are going. Here is an example list that you can use as a guide when creating very small distributions:

It is easiest to have something to start with when creating your own distribution. You can use the Yocto Project out-of-the-box to create the poky-tiny distribution. Ultimately, you will want to make changes in your own distribution that are likely modeled after poky-tiny.

Understanding some memory concepts will help you reduce the system size. Memory consists of static, dynamic, and temporary memory. Static memory is the TEXT (code), DATA (initialized data in the code), and BSS (uninitialized data) sections. Dynamic memory contains memory that is allocated at runtime, stacks, hash tables, and so forth. Temporary memory is recovered after the boot process. This memory consists of memory used for decompressing the kernel and for the __init__ functions.

To help you see where you currently are with kernel and root filesystem sizes, you can use two tools found in the Source Directory in the scripts directory:

This next tool and command helps you organize configuration fragments and view file dependencies in a human-readable form:

The root filesystem is made up of packages for booting, libraries, and applications. To change things, you can configure how the packaging happens, which changes the way you build them. You can also tweak the filesystem itself or select a different filesystem.

First, find out what is hogging your root filesystem by running the dirsize.py script from your root directory:

     $ cd <root-directory-of-image>
     $ dirsize.py 100000 > dirsize-100k.log
     $ cat dirsize-100k.log
                

You can apply a filter to the script to ignore files under a certain size. This example filters out anything below 100 Kbytes. The sizes reported by the tool are uncompressed and thus, will be smaller by a relatively constant factor in a compressed root filesystem. When you examine your log file, you can focus on areas of the root filesystem that take up large amounts of memory.

You need to be sure that what you eliminate does not cripple the functionality you need. One way to see how packages relate to each other is by using the Dependency Explorer UI with the BitBake command:

     $ cd <image-directory>
     $ bitbake -u depexp -g <image>
                

Use the interface to select potential packages you wish to eliminate and see their dependency relationships.

When deciding how to reduce the size, get rid of packages that result in minimal impact on the feature set. For example, you might not need a VGA display. Or, you might be able to get by with devtmpfs and mdev instead of udev.

Use the local.conf file to make changes. For example, to eliminate udev and glib, set the following in the local configuration file:

     VIRTUAL-RUNTIME_dev_manager = ""
                

Finally, you should consider exactly the type of root filesystem you need to meet your needs while also reducing its size. For example, consider cramfs, squashfs, ubifs, ext2, or an initramfs using initramfs. Be aware that ext3 requires a 1 Mbyte journal. If you are okay with running read-only you do not need this journal.

The kernel is built by including policies for hardware-independent aspects. What subsystems do you enable? For what architecture are you building? Which drivers do you build by default.

Run the ksize.py script from the top-level Linux build directory to get an idea of what is making up the kernel:

     $ cd <top-level-linux-build-directory>
     $ ksize.py > ksize.log
     $ cat ksize.log
                

When you examine the log, you will see how much space is taken up with the built-in .o files for drivers, networking, core kernel files, filesystem, sound, and so forth. The sizes reported by the tool are uncompressed and thus, will be smaller by a relatively constant factor in a compressed kernel image. Look to reduce the areas that are large and taking up around the "90% rule."

To examine, or drill down, into any particular area, use the -d option with the script:

     $ ksize.py -d > ksize.log
                

Using this option breaks out the individual file information for each area of the kernel (e.g. drivers, networking, and so forth).

Use your log file to see what you can eliminate from the kernel based on features you can let go. For example, if you are not going to need sound, you do not need any drivers that support sound.

After figuring out what to eliminate, you need to reconfigure the kernel to reflect those changes during the next build. You could run menuconfig and make all your changes at once. However, that makes it difficult to see the effects of your individual eliminations and also makes it difficult to replicate the changes for perhaps another target device. A better method is to start with no configurations using allnoconfig, create configuration fragments for individual changes, and then manage the fragments into a single configuration file using merge_config.sh. The tool makes it easy for you to iterate using the configuration change and build cycle.

Each time you make configuration changes, you need to rebuild the kernel and check to see what impact your changes had on the overall size.

Depending on your particular circumstances, other areas that you can trim likely exist. The key to finding these areas is through tools and methods described here combined with experimentation and iteration. Here are a couple of areas to experiment with:

If you have not reached your goals on system size, you need to iterate on the process. The process is the same. Use the tools and see just what is taking up 90% of the root filesystem and the kernel. Decide what you can eliminate without limiting your device beyond what you need.

Depending on your system, a good place to look might be Busybox, which provides a stripped down version of Unix tools in a single, executable file. You might be able to drop virtual terminal services or perhaps ipv6.

If a committed change results in changing the package output, then the value of the PR variable needs to be increased (or "bumped"). Increasing PR occurs one of two ways:

Given that one of the challenges any build system and its users face is how to maintain a package feed that is compatible with existing package manager applications such as RPM, APT, and OPKG, using an automated system is much preferred over a manual system. In either system, the main requirement is that version numbering increases in a linear fashion and that a number of version components exist that support that linear progression.

The following two sections provide information on the PR Service and on manual PR bumping.

     PRSERV_HOST = "localhost:0"
                    

     bitbake-prserv ‐‐host <ip> ‐‐port <port> ‐‐start
                    

     # It is recommended to activate "buildhistory" for testing the PR service
     INHERIT += "buildhistory"
     BUILDHISTORY_COMMIT = "1"
                    

Sometimes a package name you are using might exist under an alias or as a similarly named package in a different distribution. The OpenEmbedded build system implements a distro_check task that automatically connects to major distributions and checks for these situations. If the package exists under a different name in a different distribution, you get a distro_check mismatch. You can resolve this problem by defining a per-distro recipe name alias using the DISTRO_PN_ALIAS variable.

Following is an example that shows how you specify the DISTRO_PN_ALIAS variable:

     DISTRO_PN_ALIAS_pn-PACKAGENAME = "distro1=package_name_alias1 \
                                       distro2=package_name_alias2 \
                                       distro3=package_name_alias3 \
                                       ..."
                

If you have more than one distribution alias, separate them with a space. Note that the build system currently automatically checks the Fedora, OpenSUSE, Debian, Ubuntu, and Mandriva distributions for source package recipes without having to specify them using the DISTRO_PN_ALIAS variable. For example, the following command generates a report that lists the Linux distributions that include the sources for each of the recipes.

     $ bitbake world -f -c distro_check
                

The results are stored in the build/tmp/log/distro_check-${DATETIME}.results file found in the Source Directory.

Many pieces of software split functionality into optional modules (or plugins) and the plugins that are built might depend on configuration options. To avoid having to duplicate the logic that determines what modules are available in your recipe or to avoid having to package each module by hand, the OpenEmbedded build system provides functionality to handle module packaging dynamically.

To handle optional module packaging, you need to do two things:

     python populate_packages_prepend () {
         lighttpd_libdir = d.expand('${libdir}')
         do_split_packages(d, lighttpd_libdir, '^mod_(.*)\.so$',
                          'lighttpd-module-%s', 'Lighttpd module for %s',
                           extra_depends='')
     }
                    

     Mandatory arguments

     root
        The path in which to search
     file_regex
        Regular expression to match searched files.
        Use parentheses () to mark the part of this
        expression that should be used to derive the
        module name (to be substituted where %s is
        used in other function arguments as noted below)
     output_pattern
        Pattern to use for the package names. Must
        include %s.
     description
        Description to set for each package. Must
        include %s.

     Optional arguments

     postinst
        Postinstall script to use for all packages
        (as a string)
     recursive
        True to perform a recursive search - default
        False
     hook
        A hook function to be called for every match.
        The function will be called with the following
        arguments (in the order listed):

        f
           Full path to the file/directory match
        pkg
           The package name
        file_regex
           As above
        output_pattern
           As above
        modulename
           The module name derived using file_regex

     extra_depends
        Extra runtime dependencies (RDEPENDS) to be
        set for all packages. The default value of None
        causes a dependency on the main package
        (${PN}) - if you do not want this, pass empty
        string '' for this parameter.
     aux_files_pattern
        Extra item(s) to be added to FILES for each
        package. Can be a single string item or a list
        of strings for multiple items. Must include %s.
     postrm
        postrm script to use for all packages (as a
        string)
     allow_dirs
        True to allow directories to be matched -
        default False
     prepend
        If True, prepend created packages to PACKAGES
        instead of the default False which appends them
     match_path
        match file_regex on the whole relative path to
        the root rather than just the file name
     aux_files_pattern_verbatim
        Extra item(s) to be added to FILES for each
        package, using the actual derived module name
        rather than converting it to something legal
        for a package name. Can be a single string item
        or a list of strings for multiple items. Must
        include %s.
     allow_links
        True to allow symlinks to be matched - default
        False
                     

     PACKAGES_DYNAMIC = "lighttpd-module-.*"
                    

For RPM, IPK, and DEB package formats, it is possible to set up a repository that is a host-based package feed from which you can install packages on the target system during runtime. Doing so is optional and depends on the following:

Following are the steps to set up the optional repository. This examples assumes you are using RPM and the Apache 2 server:

After taking these steps and making sure that the other requirements mentioned at the beginning of the section are met, reboot the target device to take advantage of runtime package installations.

If your packages are IPK, you can install packages onto an existing running system by first sharing the tmp/deploy/ipk/ directory through a web server and then by changing /etc/opkg/base-feeds.conf to point at the shared server. Following is an example:

     $ src/gz all http://www.mysite.com/somedir/deploy/ipk/all
     $ src/gz armv7a http://www.mysite.com/somedir/deploy/ipk/armv7a
     $ src/gz beagleboard http://www.mysite.com/somedir/deploy/ipk/beagleboard
                

A Package Test (ptest) runs tests against packages built by the OpenEmbedded build system on the target machine. A ptest contains at least two items: the actual test, and a shell script (run-ptest) that starts the test. The shell script that starts the test must not contain the actual test, the script only starts it. On the other hand, the test can be anything from a simple shell script that runs a binary and checks the output to an elaborate system of test binaries and data files.

The test generates output in the format used by Automake:

     <result>: <testname>
                

where the result can be PASS, FAIL, or SKIP, and the testname can be any identifying string.

     DISTRO_FEATURES_append = " ptest"
     EXTRA_IMAGE_FEATURES += "ptest-pkgs"
                    

Set the following variables in your distribution configuration file as follows:

     DISTRO_FEATURES_append = " systemd"
     VIRTUAL-RUNTIME_init_manager = "systemd"
                

You can also prevent the sysvinit distribution feature from being automatically enabled as follows:

     DISTRO_FEATURES_BACKFILL_CONSIDERED = "sysvinit"
                

Doing so removes any redundant sysvinit scripts.

Set the following variables in your distribution configuration file as follows:

     DISTRO_FEATURES_append = " systemd"
     VIRTUAL-RUNTIME_init_manager = "systemd"
                

Doing so causes your main image to use the packagegroup-core-boot.bb recipe and systemd. The rescue/minimal image cannot use this package group. However, it can install sysvinit and the appropriate packages will have support for both systemd and sysvinit.

To create the read-only root filesystem, simply add the read-only-rootfs feature to your image. Using either of the following statements in your image recipe or from within the local.conf file found in the Build Directory causes the build system to create a read-only root filesystem:

     IMAGE_FEATURES = "read-only-rootfs"
                

or

     EXTRA_IMAGE_FEATURES = "read-only-rootfs"
                

For more information on how to use these variables, see the "Customizing Images Using Custom IMAGE_FEATURES and EXTRA_IMAGE_FEATURES" section. For information on the variables, see IMAGE_FEATURES and EXTRA_IMAGE_FEATURES.

It is very important that you make sure all post-Installation (pkg_postinst) scripts for packages that are installed into the image can be run at the time when the root filesystem is created during the build on the host system. These scripts cannot attempt to run during first-boot on the target device. With the read-only-rootfs feature enabled, the build system checks during root filesystem creation to make sure all post-installation scripts succeed. If any of these scripts still need to be run after the root filesystem is created, the build immediately fails. These checks during build time ensure that the build fails rather than the target device fails later during its initial boot operation.

Most of the common post-installation scripts generated by the build system for the out-of-the-box Yocto Project are engineered so that they can run during root filesystem creation (e.g. post-installation scripts for caching fonts). However, if you create and add custom scripts, you need to be sure they can be run during file system creation.

Here are some common problems that prevent post-installation scripts from running during root filesystem creation:

With the read-only-rootfs feature enabled, any attempt by the target to write to the root filesystem at runtime fails. Consequently, you must make sure that you configure processes and applications that attempt these types of writes do so to directories with write access (e.g. /tmp or /var/run).

Before you can initiate a remote debugging session, you need to be sure you have set up the cross-development environment, toolchain, and sysroot. The "Preparing for Application Development" chapter of the Yocto Project Application Developer's Guide describes this process. Be sure you have read that chapter and have set up your environment.

Make sure Gdbserver is installed on the target. If it is not, install the package gdbserver, which needs the libthread-db1 package.

Here is an example that when entered from the host connects to the target and launches Gdbserver in order to "debug" a binary named helloworld:

     $ gdbserver localhost:2345 /usr/bin/helloworld
                

Gdbserver should now be listening on port 2345 for debugging commands coming from a remote GDB process that is running on the host computer. Communication between Gdbserver and the host GDB are done using TCP. To use other communication protocols, please refer to the Gdbserver documentation.

Running GDB on the host computer takes a number of stages, which this section describes.

     /opt/poky/1.4/sysroots/i686-pokysdk-linux/usr/bin/armv7a-vfp-neon-poky-linux-gnueabi/arm-poky-linux-gnueabi-gdb
                    

     $ bitbake gdb-cross
                    

     tmp/sysroots/<host-arch>/usr/bin/<target-platform>/<target-abi>-gdb
                    

     set sysroot /home/jzhang/sysroot_arm
                    

     $ arm-poky-linux-gnuabi-gdb helloworld
                    

From the target, you need to connect to the remote GDB server that is running on the host. You need to specify the remote host and port. Here is the command continuing with the example:

     target remote 192.168.7.2:2345
                

You can now proceed with debugging as normal - as if you were debugging on the local machine. For example, to instruct GDB to break in the "main" function and then continue with execution of the inferior binary use the following commands from within GDB:

     (gdb) break main
     (gdb) continue
                

For more information about using GDB, see the project's online documentation at http://sourceware.org/gdb/download/onlinedocs/.

Using OProfile you can perform all the profiling work on the target device. A simple OProfile session might look like the following:

     # opcontrol --reset
     # opcontrol --start --separate=lib --no-vmlinux -c 5
              .
              .
        [do whatever is being profiled]
              .
              .
     # opcontrol --stop
     $ opreport -cl
                

In this example, the reset command clears any previously profiled data. The next command starts OProfile. The options used when starting the profiler separate dynamic library data within applications, disable kernel profiling, and enable callgraphing up to five levels deep.

After you perform your profiling tasks, the next command stops the profiler. After that, you can view results with the opreport command with options to see the separate library symbols and callgraph information.

Callgraphing logs information about time spent in functions and about a function's calling function (parent) and called functions (children). The higher the callgraphing depth, the more accurate the results. However, higher depths also increase the logging overhead. Consequently, you should take care when setting the callgraphing depth.

For more information on using OProfile, see the OProfile online documentation at http://oprofile.sourceforge.net/docs/.

A graphical user interface for OProfile is also available. You can download and build this interface from the Yocto Project at http://git.yoctoproject.org/cgit.cgi/oprofileui/. If the "tools-profile" image feature is selected, all necessary binaries are installed onto the target device for OProfileUI interaction. For a list of image features that ship with the Yocto Project, see the "Images" section in the Yocto Project Reference Manual.

Even though the source directory usually includes all needed patches on the target device, you might find you need other OProfile patches for recent OProfileUI features. If so, see the OProfileUI README for the most recent information.

     # opcontrol --reset
     # opcontrol --start --separate=lib --no-vmlinux -c 5
            .
            .
     [do whatever is being profiled]
            .
            .
     # opcontrol --stop
     # oparchive -o my_archive
                    

Compliance activities should begin before you generate the final image. The first thing you should look at is the requirement that tops the list for most compliance groups - providing the source. The Yocto Project has a few ways of meeting this requirement.

One of the easiest ways to meet this requirement is to provide the entire DL_DIR used by the build. This method, however, has a few issues. The most obvious is the size of the directory since it includes all sources used in the build and not just the source used in the released image. It will include toolchain source, and other artifacts, which you would not generally release. However, the more serious issue for most companies is accidental release of proprietary software. The Yocto Project provides an archiver class to help avoid some of these concerns. See the "Archiving Sources - archive*.bbclass" section in the Yocto Project Reference Manual for information on this class.

Before you employ DL_DIR or the archiver class, you need to decide how you choose to provide source. The source archiver class can generate tarballs and SRPMs and can create them with various levels of compliance in mind. One way of doing this (but certainly not the only way) is to release just the original source as a tarball. You can do this by adding the following to the local.conf file found in the Build Directory:

     ARCHIVER_MODE ?= "original"
     ARCHIVER_CLASS = "${@'archive-${ARCHIVER_MODE}-source' if
     ARCHIVER_MODE != 'none' else ''}"
     INHERIT += "${ARCHIVER_CLASS}"
     SOURCE_ARCHIVE_PACKAGE_TYPE = "tar"
                

During the creation of your image, the source from all recipes that deploy packages to the image is placed within subdirectories of DEPLOY_DIR/sources based on the LICENSE for each recipe. Releasing the entire directory enables you to comply with requirements concerning providing the unmodified source. It is important to note that the size of the directory can get large.

A way to help mitigate the size issue is to only release tarballs for licenses that require the release of source. Let's assume you are only concerned with GPL code as identified with the following:

     $ cd poky/build/tmp/deploy/sources
     $ mkdir ~/gpl_source_release
     $ for dir in */*GPL*; do cp -r $dir ~/gpl_source_release; done
                

At this point, you could create a tarball from the gpl_source_release directory and provide that to the end user. This method would be a step toward achieving compliance with section 3a of GPLv2 and with section 6 of GPLv3.

One requirement that is often overlooked is inclusion of license text. This requirement also needs to be dealt with prior to generating the final image. Some licenses require the license text to accompany the binary. You can achieve this by adding the following to your local.conf file:

     COPY_LIC_MANIFEST = "1"
     COPY_LIC_DIRS = "1"
                

Adding these statements to the configuration file ensures that the licenses collected during package generation are included on your image. As the source archiver has already archived the original unmodified source that contains the license files, you would have already met the requirements for inclusion of the license information with source as defined by the GPL and other open source licenses.

At this point, we have addressed all we need to address prior to generating the image. The next two requirements are addressed during the final packaging of the release.

By releasing the version of the OpenEmbedded build system and the layers used during the build, you will be providing both compilation scripts and the source code modifications in one step.

If the deployment team has a BSP layer and a distro layer, and those those layers are used to patch, compile, package, or modify (in any way) any open source software included in your released images, you may be required to to release those layers under section 3 of GPLv2 or section 1 of GPLv3. One way of doing that is with a clean checkout of the version of the Yocto Project and layers used during your build. Here is an example:

     # We built using the dylan branch of the poky repo
     $ git clone -b dylan git://git.yoctoproject.org/poky
     $ cd poky
     # We built using the release_branch for our layers
     $ git clone -b release_branch git://git.mycompany.com/meta-my-bsp-layer
     $ git clone -b release_branch git://git.mycompany.com/meta-my-software-layer
     # clean up the .git repos
     $ find . -name ".git" -type d -exec rm -rf {} \;
                

One thing a development organization might want to consider for end-user convenience is to modify meta-yocto/conf/bblayers.conf.sample to ensure that when the end user utilizes the released build system to build an image, the development organization's layers are included in the bblayers.conf file automatically:

     # LAYER_CONF_VERSION is increased each time build/conf/bblayers.conf
     # changes incompatibly
     LCONF_VERSION = "6"

     BBPATH = "${TOPDIR}"
     BBFILES ?= ""

     BBLAYERS ?= " \
       ##COREBASE##/meta \
       ##COREBASE##/meta-yocto \
       ##COREBASE##/meta-yocto-bsp \
       ##COREBASE##/meta-mylayer \
       "

     BBLAYERS_NON_REMOVABLE ?= " \
       ##COREBASE##/meta \
       ##COREBASE##/meta-yocto \
       "
                

Creating and providing an archive of the Metadata layers (recipes, configuration files, and so forth) enables you to meet your requirements to include the scripts to control compilation as well as any modifications to the original source.